AMD Radeon AI PRO R9700 GPU announced, RDNA 4 and 32GB of memory

The new AMD Radeon AI PRO R9700 GPU has been announced at Computex 2025, bringing RDNA 4 to AI-powered workstations. With 32GB of GDDR6 memory, 96 TFLOPS of Peak Half-Precision, and 1531 TOPS of INT4 Sparse AI performance, you're looking at up to 2X better performance than the previous generation's AMD Radeon PRO W7800 32GB GPU.

4

With 32GB of VRAM, AMD notes that the Radeon AI PRO R9700 is better equipped for running advanced local text-to-image AI models and LLMs like DeepSeek R1 Distill Qwen 32B Q6 and Mistral Small 3.1 24B Instruct 2503 Q8. With expanded AMD ROCm on Radeon, there'll be support for a "broader range of AI and compute workloads."

The first chart in the presentation for the new AI workstation GPU compares the average tokens per second performance against the GeForce RTX 5080, showing an increase of up to 496%. Check out the chart below.

4

The Radeon AI PRO R9700 is also scalable. Up to four GPUs in a single workstation offer enough memory to handle 123 billion and 70 billion parameter models, which is some serious AI performance. AMD notes that the Radeon AI PRO R9700 will launch in July 2025 (pricing is TBC), with several models from its partners - ASRock, ASUS, GIGABYTE, PowerColor, Sapphire, XFX, and Yeston - set to become available.

4